Scalable Authoritative OWL Reasoner Aidan Hogan, Andreas Harth, Axel Polleres

1. Scalable Authoritative OWL Reasoner Aidan Hogan, Andreas Harth, Axel Polleres Digital Enterprise Research Institute National University of Ireland, Galway =“free” (Irish)

2. SAOR - Reasoning for SWSE http://swse.deri.org/ • We want the challenge data plus OWL inferred data in the search results! • Our approach: SAOR – Scalable Authoritative OWL Reasoning

3. Idea • Apply a subset of OWL reasoning to the billion triple challenge dataset • Forward-chaining rule based approach, e.g.[ter Horst, 2005] • Reduced output statements for the SWSE use case… • Must be scalable, must be reasonable • … incomplete w.r.t. OWL BY DESIGN! • SCALABLE: Tailored ruleset • file-scan processing • avoid joins • AUTHORITATIVE: Avoid Non-Authoritative inference (“hijacking”, “non-standard vocabulary use”)

4. Scalable Reasoning • Scan 1: Scan all data (1.1b statements), separate T-Box statements, load T-Box statements (8.5m) into memory, perform authoritative analysis. • Scan 2: Scan all data and join all statements with in-memory T-Box . • Only works for inference rules with 0-1 A-Box patterns • No T-Box expansion by inference  Needs “tailored” ruleset

5. Rules Applied: Tailored version of [ter Horst, 2005]

6. Good “excuses” to avoid G2 rules • The obvious: • G2 rules would need joins, i.e. to trigger restart of file-scan • The interesting one: • Take for instance IFP rule: • Maybe not such a good idea on real Web data • More experiments including G2, G3 rules in [Hogan, Harth, Polleres, ASWC2008]

7. Authoritative Reasoning • Document D authoritative for concept C iff: • C not identified by URI • OR • De-referenced URI of C coincides with or redirects to D • FOAF spec authoritative for foaf:Person✓ • MY spec not authoritative for foaf:Person✘ • Only allow extension in authoritative documents • my:Person rdfs:subClassOf foaf:Person . (MY spec) ✓ • BUT: Reduce obscure memberships • foaf:Person rdfs:subClassOf my:Person . (MY spec) ✘ • Similarly for other T-Box statements. • In-memory T-Box stores authoritative values for rule execution Ontology Hijacking

8. Rules Applied The 17 rules applied including statements considered to be T-Box, elements which must be authoritatively spoken for (including for bnode OWL abstract syntax), and output count

9. Authoritative Resoning covers rdfs: owl: vocabulary misuse • http://www.polleres.net/nasty.rdf: rdfs:subClassOf rdfs:subPropertyOf rdfs:Resource. rdfs:subClassOf rdfs:subPropertyOf rdfs:subPropertyOf. rdf:type rdfs:subPropertyOf rdfs:subClassOf. rdfs:subClassOf rdf:type owl:SymmetricProperty. • Naïve rules application would infer O(n3) triples • By use of authoritative reasoning SAOR/SWSE doesn’t stumble over these  :rdfs :owl Hijacking

10. Performance Graph showing SAOR’s rate of input/output statements per minute for reasoning on 1.1b statements: reduced input rate correlates with increased output rate and vice-versa

11. Results • SCAN 1:6.47 hrs • In-mem T-Box creation, authoritative analysis: • SCAN 2:9.82 hrs • Scan reasoning – join A-Box with in-mem authoritative T-Box: • 1.925b new statements inferred in 16.29 hrs • On our agenda: • More valuable insights on our experiences from Web data • G2 and G3 rules? • Detailed comparison to OWL RL 1.1b + 1.9b inferred = 3 billion triples in SWSE

12. Search result example:

13. Le Fin… Enjoy the data… GUI: http://swse.deri.org/ SPARQL interface: http://swse.deri.org/yars2/ Contact us for feedback!